minor fix

CHANGELOG.md

@@ -6,6 +6,8 @@
 
 ### code
 
+- fixed type inference for `backend.from_numpy()` for torch backend
+
 - fixed typo in function name (read_bruker_best3_dataset instead of
   read_bruker_bes3t_dataset), old name was kept available for backward
   compatibility
@@ -15,6 +17,8 @@
 
 ### documentation
 
+- fixed bibtex reference [Bar21]
+
 - fixed minor issues in demonstration examples
 
 - fixed pip installation instructions in [README.md](README.md)

docs/references.bib

@@ -75,7 +75,7 @@ @article{Barnett_2019
 }
 
 @article{Barnett_2021,
-	title = {Aliasing error of the exp⁡(β1−z2) kernel in the nonuniform fast Fourier transform},
+	title = {Aliasing error of the exp(β·sqrt(1-z^2)) kernel in the nonuniform fast Fourier transform},
 	journal = {Applied and Computational Harmonic Analysis},
 	volume = {51},
 	pages = {1-16},

src/pyepri/backends.py

@@ -711,8 +711,8 @@ def __init__(self, lib, device):
             self.erfc = lambda x, out=None : lib.erfc(x, out=out)
             self.is_complex = lambda x : x.is_complex()
             self.to_numpy = lambda x : x.detach().cpu().numpy()
-            #self.from_numpy = lambda x : lib.from_numpy(x).to(device, copy=True) # confilct between pytorch and numpy 2.0.0
-            self.from_numpy = lambda x : lib.Tensor(x).to(device, copy=True)
+            #self.from_numpy = lambda x : lib.from_numpy(x).to(device, copy=True) # conflict between pytorch and numpy 2.0.0
+            self.from_numpy = lambda x : lib.Tensor(x).to(device, copy=True, dtype=self.str_to_lib_dtypes[str(x.dtype)]) # need explicit type
             self.cast = lambda x, dtype : x.type(self.str_to_lib_dtypes[dtype])
             self.transpose = lambda x : x.moveaxis((0,1),(1,0))
             self.copy = lambda x : lib.clone(x).detach()
@@ -782,14 +782,18 @@ def __init__(self, lib, device):
             self.is_complex.__doc__ = "return x.is_complex()"
             self.to_numpy.__doc__ = "return x.detach().cpu().numpy()"
             #self.from_numpy.__doc__ = "return torch.from_numpy(x).to('" + self.device + "', copy=True)"
-            self.from_numpy.__doc__ = "return torch.Tensor(x).to('" + self.device + "', copy=True)"
+            self.from_numpy.__doc__ = (
+                "return torch.Tensor(x).to('" + self.device + "', copy=True, dtype=self.str_to_lib_dtypes[str(x.dtype)])\n"
+                "where `self` denotes the backends.Backend class instance from wich this lambda\n"
+                "function belongs to."
+            )
             self.is_backend_compliant.__doc__ = "return all([isinstance(arg, torch.Tensor) for arg in args])"
             self.quantile.__doc__ = "return "+ lib.__name__ + ".quantile(u, q, dim=dim, keepdim=keepdim, out=out, interpolation=interpolation)"
             self.frombuffer.__doc__ = (
                 "return " + lib.__name__ + ".frombuffer(buffer, dtype=self.str_to_lib_dtypes[dtype], count=count, offset=offset)\n"
                 "where `self` denotes the backends.Backend class instance from wich this lambda\n"
                 "function belongs to."
-                )
+            )
 
             # nufft support (use finufft for CPU device and cufinufft
             # for GPU device)